You have 1 article left to read this month before you need to register a free LeadDev.com account.
Estimated reading time: 3 minutes
A new study from Multitudes identifies 4 tactics to get the most out of AI rollouts.
As organizations race to deploy artificial intelligence tools, new research suggests the biggest determinant of success isn’t the technology itself, but how leaders guide its adoption.
Engineering analytics firm Multitudes collected data from more than 500 developers across multiple AI rollouts. In the resulting AI impact whitepaper, the researchers concluded that simply introducing AI tools does not automatically lead to efficiency gains. Instead, it highlights a set of leadership practices that consistently separate effective rollouts from unsuccessful ones.
“The bad news is that buying AI tooling doesn’t guarantee good outcomes – or even decent AI adoption,” the paper states. “The good news is that what does make a big difference is what you do as a leader. The culture and expectations you set for your people seems to be what makes the difference between a successful AI rollout and a lackluster one.”
Clarity over hype
When it comes to AI engagement, engineering leaders are often caught between pressure from above and uncertainty on the ground.
In this situation, it is down to leaders to provide as much clarity as possible to counter any exaggerated expectations.
Without clear goals and guidance, it’s normal for teams to struggle to understand how AI is meant to support their work, leading to inconsistent adoption and missed opportunities.
“Make sure the “why” of your AI rollout is clear, and repeat it until everyone on your team knows it by heart,” the paper stated.
Your inbox, upgraded.
Receive weekly engineering insights to level up your leadership approach.
Delivery pressure can stall adoption
Even when AI tools are available, intense delivery pressure can slow adoption, particularly among senior engineers.
Breaking the Multitudes data down by seniority revealed a clear divide: junior and intermediate developers increased their AI use during the high-pressure period, while senior engineers largely paused adoption, reverting to old ways of working. That senior adoption rebounded once the launch was complete.
In contrast, junior and intermediate engineers used AI even under pressure to fill knowledge gaps, navigate unfamiliar code, and work more efficiently with team resources.
“In our interviews, we saw a common concern from senior engineers about the hype of AI, and skepticism about whether the productivity impact of AI would be as big as the headlines claim,” the paper explained.
The lesson? Engineers won’t have as much time to learn when they’re facing significant delivery pressures, so expectations need to be adjusted accordingly.
More like this
Don’t forget code quality
Speed is one thing, but code quality also needs to be accounted for when introducing AI tools and agents. The research identifies two early indicators of code quality that leaders should monitor:
- Pull request (PR) size: Larger PRs are harder to review, more error-prone, and slow down collaboration. AI-generated code often increases PR size, which can signal declining quality.
- Human review quantity and quality: Developers remain the primary safeguard for code health. High-quality human reviews catch errors that AI might introduce, especially in complex changes.
“When leaders set code quality as a goal and have strong code review norms, they can overcome the quality issues of AI,” the paper states. “Getting leading indicators of how AI is impacting code quality can help leaders stay ahead of issues and make sure their AI interventions are taking the organization in the desired direction.”
Peer-to-peer sharing
Successful AI adoption often comes down to how top performers use the tools, not just having the tools available. The research highlights the importance of identifying your organization’s AI super-users and creating structured ways for them to share their practices. This could be through an AI guild, or dedicated learning sessions.
“Find your super-users and see what they do differently,” the report recommends. “Document the specific prompts, workflows, and review practices that work on your codebase. Support peer-to-peer learning: Host regular sessions where high-performing super-users demo their work.
“Developers trust their peers, and seeing AI work on their actual codebase and their specific problems accelerates adoption.”
London • June 2 & 3, 2026
Rands, Nicole Forsgren & Matej Pfajfar confirmed
